Memo - Columbia University

CCB-PRO-02/04 Rev.1

on-

Memo

To: CCB

From: Paul Richards, Relu Burlacu, and Mark Fisk (Columbia University Group 1 Csortium)

Date: May 17, 2002

Subject: Pn SSSCs for IMS and Surrogate Stations in Asia

Sponsor: Robert Woodward

CC:

Abstract

We propose that Pn Source Specific Station Corrections (SSSCs) for 14 seismic stations in Asia

be installed into the CMR baseline processing configuration. The SSSCs are expected to improve

the accuracy of location estimates of seismic events in Asia and reduce the uncertainties. The

SSSCs were computed by the method of Bondár (1999) using regionalization (Figure 1) of 1-D

travel-time curves versus distance, as described in detail in Appendix A. These model-basedSSSCs were then refined by a kriging algorithm (e.g., Bottone et al., 2001), in which the SSSCs

are updated at each gridpoint by a linear combination of travel-time residuals for ground-truth

(GT) events, weighted by a distance-dependent correlation function. The kriging algorithm also

estimates an uncertainty grid. For well-calibrated areas, the correction grid converges to the mean

of local data and the uncertainty converges to the residual variance of local data. Far from calibra-

tion data, the correction grid converges to the model-based SSSC value, with larger uncertainty

equal to the sum of the residual and calibration variances, which is the variance of the travel-time

means averaged over all well-separated locations.

Several tests were conducted on the RDSS testbed to validate the SSSCs, with emphasis on model

validation and evaluation of the kriged SSSCs. Two data sets were used for the validation: Kitov’s

data set of phase picks for nuclear weapons tests at Semipalatinsk and Peaceful Nuclear Explo-

sions (PNEs) recorded by various combinations of 93 stations, available through Harvard’s web

site (www.seismology.havard.edu/~ekstrom/Research/FSU_data/FSU_data.html); and a second

data set of 18 nuclear explosions in China, India, and Pakistan, and chemical explosions at Semi-

palatinsk recorded by a sparse network of IMS seismic stations. A more detailed description of

the data sets is provided in Appendix A. Using Pn arrival times assembled by Kitov at the IDG in

Moscow, and published GT locations and origin times, we relocated 156 events recorded by vari-

ous combinations of 93 regional stations, with and without using SSSCs. Mislocations are

of 20

CCB-PRO-02/04 Rev.1

ults fortinsk,n data

reduced for 63% of the events using the model-based SSSCs, and for 93% using the kriged

SSSCs. The median mislocation improved from 12.2 km to 9.5 km and 2.7 km, respectively. The

median area of the error ellipses was reduced from 1,596 km2 to 450 km2 and 196 km2, respec-

tively. Error ellipse coverage (percent of GT event locations within the corresponding error

ellipses) is 97% without using SSSCs, 94% using model-based SSSCs, and 100% using kriged

SSSCs. These results were obtained for source locations, stations, and paths that sample very

extensive and diverse geological provinces of Central and Northern Asia. Thus, we believe the

results indicate the general validity of the model and the resulting SSSCs for this region. Weexpect these SSSCs to perform, on average, as well as indicated by the validation test resthe model-basedSSSCs, and substantially better for areas near the Lop Nor, SemipalaIndian and Pakistani nuclear test sites, and sites of historical Soviet PNE’s, where calibratiohave been utilized.

Statement of Objective

The objective is to install Pn SSSCs, computed by a regionalized travel-time model and kriging,

for 14 IMS and surrogate stations in Asia into the CMR baseline processing configuration. These

surface-focus (i.e., for depth = 0 km) SSSCs improve seismic event location performance in Asia

by reducing mislocations and error ellipse areas while achieving at least 90% coverage.

Summary of Proposed Change

The proposed change is to install, under the CMR version control system (ClearCase), the Pn

SSSCs for 14 IMS and surrogate stations in Asia, presented in Table 1, including stations AAK,

AKTO, BRVK, KURK, MAKZ, NIL, ZAL, MAG, NRI (NRIS), SEY, TIK (TIXI), TLY, YAK,

ULN (JAVM). Plots of the SSSCs and the corresponding modeling errors are presented in Burlacu

et al. (2002).

Note that SSSCs for seven of these stations have already been delivered by the SAIC Group 2

Consortium. Although preliminary comparisons indicate that our Pn SSSCs for these stations per-

form significantly better for GT events in Asia, further work is needed to devise a strategy and

thoroughly compare location performance using these duplicate sets of SSSCs. This evaluation

work should be performed before these SSSCs are used in the operational system.

Expected Benefits

The off-line tests performed on the RDSS testbed and by the Columbia University Group 1 con-

sortium demonstrate significant improvements in all location performance metrics recommended

at the IDC Technical Experts Meeting on Seismic Event Location in 1999 (CTBT/WGB/TL-

2/18). The methods of computing the SSSCs, the regionalization of 1-D travel-time curves, the

GT data, and the results of testing are thoroughly documented by Burlacu et al. (2002). Validation

tests performed by RDSS staff are documented by Bahavar (2002). Based on these validation test

results for GT events in extensive and diverse provinces of Asia, we fully expect these Pn SSSCs

to provide comparable improvements in location performance for the operational system.

of 20

CCB-PRO-02/04 Rev.1

m (tri-own onet thattium isents insionsade

ents ingions,t Asia,

Possible Risks and Dependencies

The SSSCs have potential impacts on the performance of GA, which uses libloc, and on EvLoc,

which is called by automatic processing and ARS. Although we have not performed tests to assess

the impact of our SSSCs on GA and ARS, the CCB Memo by Bondár (2001) addresses this issue.

Their conclusion was that the impact of SSSCs, in general, is not considerable in terms of compu-

tational load and memory requirements. One possible issue is that larger SSSC files require

increased read-in time. The SSSCs, computed by Columbia University Group 1 consortium, were

generated on a rectangular grid with one degree spacing in longitude and latitude. As this is the

same format as for previous SSSCs, we do not anticipate any risks of using the SSSCs in the GAand ARS environments. The off-line tests performed by RDSS staff and us showed no adverse

impact of the SSSCs on EvLoc. However, we recommend that basic on-line testing be performed

before the proposed Pn SSSCs are installed into the operational environment.

Note that not all events are affected by the proposed changes, but only those shallow events

recorded by the 14 seismic stations in Asia with Pn phases.

Summary of Testing

Off-line testing was performed by the Columbia University Group 1 consortium and indepen-

dently by RDSS staff. Appendix A and Appendix C provide the detailed results as Validation Test

Reports from the Columbia University Group 1 consortium (Burlacu et al., 2002)and the RDSS

testbed (Bahavar, 2002), respectively. Here we summarize the main results of those tests.

Figure 2 shows a map of the GT event locations (red stars) and the stations that recorded theangles), along with great circle paths between events and stations. There are 174 events shthe map, including 156 explosions in Kitov’s data set and 18 explosions in a separate data swe have assembled. The green triangles represent the 30 IMS stations for which our consortasked with generating SSSCs and the smaller blue triangles are stations that recorded evKitov’s data set. The green curves in Figure 2 indicate the great circle paths for the 18 explofrom the second data set, along with paths for PNE’s recorded by BRVK for which we have mour own phase picks from waveforms. The blue curves are great circle paths between evKitov’s data set and the various 93 recording stations. This map illustrates that the source restation sites, and paths sample very diverse and extensive geological structures throughoumaking this data set extremely valuable for model validation.

The main objective of these tests is to use GT events to validate the Pn SSSCs by evaluating:

• travel time residuals before and after corrections;

• location performance (mislocation errors, error ellipse size, and coverage).

Figure 3 shows travel-time residuals versus epicentral distance for 145 Pn arrivals at station

BRVK, corresponding to Soviet-era PNE’s and UNE’s at the Semipalatinsk and Lop Nor nuclear

test sites. The residuals are relative to IASPEI91 without corrections (red squares) and after

applying the model-based SSSCs computed by Bondár’s method (green circles). It is clear the

model-based SSSCs generally reduce the travel-time residuals. Figure 4 shows a similar plot

of 20

CCB-PRO-02/04 Rev.1

l andodel

n theby the

using the SSSCs computed by Bondár’s method and kriging, which shows that application of

kriging provides further reduction of the Pn travel-time residuals at BRVK.

These results are quantified in Table 2 in terms of the mean and standard deviation of the Pn

travel-time residuals for the various sets of explosions and the overall results. In all cases, both the

mean travel-time bias and the standard deviation of the travel-time residuals are progressively

reduced by applying the model-based SSSCs and the model-based plus kriged SSSCs. Although

these results for BRVK were shown because of the geographical distribution of GT events and the

quantity and quality of the Pn phase picks, which we carefully reviewed by inspection of the

waveforms, they are qualitatively representative of the reductions in mean travel-time bias and

residual variance that we obtain at other stations in Asia.

To evaluate the location performance we looked at two key aspects:(1) model validation and (2)evaluation of the kriged SSSCs. The first is to validate the regionalized travel-time modemodel-basedSSSCs computed by Bondár’s method. The goal is to demonstrate that this mprovides an effective representation of travel times in Central Asia. This is a critical step ivalidation process because events may occur at locations far from calibration points usedkriging algorithm, where the grids are asymptotically equivalent to themodel-based SSSCs.

The second goal is to assess the location performance using the kriged SSSCs. To do this, we relo-

cated the GT events using the kriged SSSCs with a leave-one-out procedure (to avoid using the

same events to both compute and test the grids) and quantify the results in terms of the same per-

formance metrics used in the model validation. The results are compared to those in which the

relocations were performed without SSSCs and with the SSSCs computed by Bondár’s method.

For model validation we used 156 events recorded at 93 stations (Kitov’s data set). The test con-

sists of relocating these events using Pn arrivals (2626 picks). All the relocations are performed

with depth fixed at the surface. The relocation procedure is first applied using the IASPEI91

travel-time tables, without any SSSCs. This is followed by relocating the same events using the

SSSCs. Both sets of relocation results are saved in a database at the CMR. Executing EvLoc with

and without SSSCs resulted in 156 events with location estimates that converged.

Relocation performance is quantified using metrics that conform with the guidelines from the

1999 Location Workshop (CTBT/WGB/TL-2/18; also provided in Appendix D) held in Oslo,

Norway, which include the following:

• the median mislocation of GT events should be significantly reduced;

• mislocation should be reduced by 20% or more for the majority of events;

• median area of confidence ellipses should be reduced, and the coverage should be the same or

better;

• confidence ellipses should be reduced by 20% or more for the majority of events;

• variance of travel-time residuals should be similar or smaller.

of 20

CCB-PRO-02/04 Rev.1

Mislocation using Model-Based SSSCs

Mislocation is expressed as the distance between the GT location and the location obtained by

EvLoc. Of the 156 events, the locations using SSSCs improved for 99 events (63%) and deterio-

rated for 57 events (37%). The median mislocation was reduced from 12.2 km to 9.5 km. For 82

events (53%) the solutions improved by more than 20%, while for 37 events (24%) the deteriora-

tion is more than 20%. Figure 5 shows the mislocation results. The green symbols represent the

events for which the relocation with SSSCs is closer to the GT location than without SSSCs. The

red symbols show events for which the mislocation without SSSCs is smaller than with SSSCs.

Error Ellipse Area and Coverage using Model-Based SSSCs

The error ellipses have systematic reduction in area by using the SSSCs than not. The difference

in the error ellipse calculations for the two cases is due to a difference in the modeling errors.

Since the modeling error for the SSSCs is always less than for IASPEI91, we expect the error

ellipses for the SSSC case to always be smaller than for the IASPEI91 case. In fact, all 156 solu-

tions (100%) are improved by more than 20% (Figure 6). The decrease in the median error ellipse

area is 1,146 km2 (from 1,596 km2 to 450 km2).

Error ellipse coverage is defined as the percentage of GT event locations that fall within the corre-

sponding 90%-confidence error ellipse. For relocation solutions without using SSSCs, 151 events

(97%) have 90%-confidence ellipses contain the GT locations. Using SSSCs, 146 events (94%)

have 90%-confidence ellipses that contain the GT locations. Although the coverage is slightly

lower when using the SSSCs, in both cases they are above the target of 90%, while the median

area of the error ellipses is reduced substantially for all the events relocated with SSSCs.

The relocation results using the model-based SSSCs show the following:

• 63% of the events are located closer to the GT location than without using SSSCs;

• error ellipse area is smaller by 20% or more for 100% of the events;

• the coverage of the error ellipses is better than 90%.

Given the large number of source regions, stations, and ray paths that sample very diverse and

extensive geological structures (represented by the 25 regions with corresponding travel times),

we expect that SSSCs computed by Bondár’s method for other stations in the same general area of

Asia will, on average, perform as well as for the stations used to compile these evaluation metrics.

Mislocation using Kriged SSSCs

To evaluate location performance using the kriged SSSCs, we use a “leave-one out” procedure in

which the event to be relocated is excluded from the kriging calculation of the SSSCs. We then

relocate each of the 156 events with kriged SSSCs that are re-computed for each event so that the

same data are not used to both compute and test the SSSCs. Since the kriged SSSCs approach the

model-based SSSCs far from calibration data, we expect that the kriged SSSCs should perform at

least as well as the model-based SSSCs, and much better for areas close to calibration data.

of 20

CCB-PRO-02/04 Rev.1

rs

Of the 156 GT events, 145 solutions (93%) have smaller mislocation errors using kriged SSSCs

than those obtained using just the IASPEI91 travel-time tables. Of these, 139 events (89%) have

mislocation errors that are reduced by more than 20%. Only 11 solutions (7%) deteriorated, but

not dramatically. The median mislocation is reduced from 12.2 km to 2.7 km when kriged SSSCs

are used. Figure 7 shows a scatter plot of the mislocation distances, relative to the GT locations,

obtained with (x-axis) and without (y-axis) using the SSSCs. As in Figure 4, the green symbols

represent events for which location estimates using the kriged SSSCs are closer to the GT loca-

tions, while the red symbols show solutions that are better without using SSSCs.

Error Ellipse Area and Coverage using Kriged SSSCs

Using kriged SSSCs, error ellipse area is reduced for 153 of 156 solutions (98%), 152 of which

(97%) are improved by more than 20%. Only 3 solutions (2%) do not have smaller error ellipses.

The median ellipse area is reduced from 1,596 km2 to 196 km2. The results are shown in Figure 8.

Error ellipse coverage, computed as the percentage of GT event locations contained within the

90%-confidence error ellipses, is 100% (all 156 GT events) when using the kriged SSSCs, as

compared to 97% (151 GT events) without using SSSCs (i.e., using IASPEI91 only).

The relocation results using kriged SSSCs show significant improvements for all location perfor-

mance metrics. Specifically,

• 93% of the events are located closer to the GT location with median mislocation erroreduced from 12.2 km to 2.7 km;

• error ellipse area is reduced by 20% or more for 97% of the events;

• median error ellipse area is reduced from 1,596 km2 to 196 km2, while achieving 100%coverage of the error ellipses with the GT event locations.

Location Results for 18 Additional GT Explosions

To directly evaluate the Pn SSSCs for IMS stations, we relocated 18 GT explosions in China,

Kazakhstan, India, and Pakistan. Although this data set is small, comparable reductions in mislo-

cations and error ellipse areas were obtained, as for the tests against Kitov’s data set (see Appen-

dix A). Note that although these events were recorded regionally by a sparse set of IMS stations,

the results are comparable to those obtained using a larger network of 93 stations for Kitov’s data

set. In all cases, the test results demonstrate that our regionalization of 1-D travel-times curves,

along with the computational methods of Bondár (1999) and kriging, have produced Pn SSSCs

and modeling errors that improve the performance of location and uncertainty estimates in Asia.

General Remarks

It is important to note that location performance for events in areas far from existing calibration

data should be, on average, comparable to the results obtained using the model-based SSSCs (i.e.,

without kriging). To illustrate the impact of kriging the travel-time residuals relative to the model-based SSSCs, we present in Figures 9-11 the kriged surfaces for stations BRVK, NRI, and YAK.

In some areas the residuals are small and kriging has little impact, indicating that the model-basedSSSCs are suitable. However, kriging can improve the SSSCs by up to several seconds.

of 20

CCB-PRO-02/04 Rev.1

rri-

using

E.and

ent,

Figure 12 shows Pn modeling errors as functions of epicentral distance for IASPEI91, our region-

alized model (green curve), and three types of regions of Northern Eurasia (platform areas, paleo-

zoic massifs and young platform, and tectonically active regions) defined by Kirichenko and

Kraev (2001). Modeling errors were calculated as standard deviations of empirical data from the

estimated travel-time curves in a 2-degree moving window with a 50% overlap (Kirichenko and

Kraev, 2001). As an initial hypothesis, we defined our Pn modeling error as an upper bound of the

modeling errors estimated by Kirichenko and Kraev (2001) for the various geotectonic provinces.

To test the validity of our Pn model error, we computed average absolute travel-time misfits to the

model-based SSSCs, binned by distance, for Pn phase arrivals in Kitov’s data set (blue circles in

Figure 12). While there are slight differences between the average misfit values and our modeling

error curve, they are not significant with respect to the uncertainties. Furthermore, the validity of

the error model was ultimately demonstrated by achieving appropriate coverage statistics. In the

future we plan to refine this curve and, if data allow, to develop region-dependent modeling errors.

Schedule and Plan for Implementation

We recommend that the 14 Pn SSSCs be installed under CMR version control system (ClearCase)

at the convenience of CMR staff. The SSSCs, in standard IDC format, are stored on the RDSS

testbed under /home/rdtst/PROJECTS/G1/1/Install/SSSC_delivery. The naming convention of the

files are TT.$sta.$phase.reg.easia.

Costs and Resources Required for Implementation

Implementation of the proposed changes is estimated to require less than one day of labor. No

other costs or resources are required.

References

Bahavar, M. RDSS Validation Test Report May 2, 2002.

Bondár, I. Application of SSSCs in the automatic processing,CCB Memo, 2001

Bondár, I. Combining 1-D models for regional calibration, inProceedings of a Workshop on IMSLocation Calibration, Oslo, January 1999.

Kirichenko, V. V., and Y. A. Kraev. Results of 1-D location calibration studies related to the tetory of Northern Eurasia,Proceedings of the 23rd Seismic Research Review, 2001.

Bottone, S., M. D. Fisk, and G. D. McCartor. Regional seismic event characterization aBayesian formulation of simple kriging, in press,Bull. Seism. Soc. Am., 2001.

Burlacu, V., M. Fisk, V. I. Khalturin, W-Y. Kim, P. G. Richards, J. Armbruster, I. Morozov,Morozova. Validation Test Report: Off-line Testing of a Regionalized Travel-Time ModelSource-Specific Stations Corrections for Asia., 2002.

Working Group B, Recommendations for seismic event location calibration developmCTBT/WGB/TL-2/18, 1999.

of 20

CCB-PRO-02/04 Rev.1

Table 1. IMS Stations for which Pn SSSCs were computed

IMS Code Country Station Name Station Code

PS23 Kazakhstan Makanchi MKAR

PS25 Mongolia Javhlant JAVM

PS29 Pakistan Pari NIL

PS33 Russian Federation Zalesovo ZAL

PS34 Russian Federation Norilsk NRIS

AS57 Kazakhstan Borovoye BRVK

AS58 Kazakhstan Kurchatov KURK

AS59 Kazakhstan Aktyubinsk AKTO

AS60 Kyrgyzstan Ala-Archa AAK

AS86 Russian Federation Seymchan SEY

AS87 Russian Federation Talaya TLY

AS88 Russian Federation Yakutsk YAK

AS91 Russian Federation Tiksi TIXI

AS93 Russian Federation Magadan MAG

Table 2. Comparison of Pn travel-time residuals for station BRVK.

Case IASPEI91Model-Based

SSSCsModel + Kriged

SSSCs

µ∆Τ σ∆Τ µ∆Τ σ∆Τ µ∆Τ σ∆Τ

Semipalatinsk UNE’s 0.51 0.45 0.11 0.44 -0.02 0.30

Soviet PNE’s -3.91 1.96 -0.51 1.35 -0.05 1.09

Lop Nor UNE’s -2.52 0.04 0.85 0.04 0.02 0.04

Overall -1.56 2.56 -0.15 1.01 -0.02 0.76

of 20

CCB-PRO-02/04 Rev.1

Figure 1. Map of topography and regionalization boundaries.

of 20

CCB-PRO-02/04 Rev.1

se

in

Figure 2. Map showing locations of 174 GT explosions (red stars) and the recording seismographic station(green triangles) used for validation tests. Also shown are great circle paths between events and stations. Thgreen triangles represent the 30 IMS stations for which our consortium is tasked with generating SSSCs. Thesmaller blue triangles represent stations that recorded events in Kitov’s data set. The green curves indicate thegreat circle paths for the 18 explosions from our second data set, and for PNE’s recorded by BRVK for whichwe have made our own phase picks from waveforms. The blue curves are great circle paths between eventsKitov’s data set and the various 93 recording stations.

30˚ 40˚ 50˚ 60˚ 70˚ 80˚ 90˚ 100˚ 110˚ 120˚ 130˚ 140˚ 150˚ 160˚

10˚

20˚

30˚

40˚

50˚

60˚

70˚

0 1000 2000

km

ULN

NIL

AAAAABABSAKH

ALUANN

ANR

APA

ARU

ASHBAK

BEYBKR

BOD

BRV

CHM

CITCUR

DSH

DUS

ELT

ERE

FABFEO

FRG

FRU

GAR

GEGGORGRO

GRS

ILI

IRKKAB

KAT

KDS

KHO

KPC

KRMKRS

KRV

KUL

KYA

LENLNK

MA1MAG

MAK

MIKMOS

MOY

MURNAKNAM

NRI

NRN

NVSOBN

PRLSPRZ

PUL

PYARYB

SAM

SEM

SEY

SHESHKS

SIMSOC

STE

SVE

TASTI2

TIK

TUP

UERUSK

VAN

YAK

YAL

ZAK

ZEIZUG

ABKT

HIA

LZH

MKAR JAVM

PRPKKSAR

ZAL

NRIS

PDYAR

USK

CMARCHT

BJT

KMI

SSEXAN

BRVKKURKAKT0

AAK

EVN

SEY

TLY

YAK

URG

TIXI

YSS

MAG

COC

of 20

CCB-PRO-02/04 Rev.1

Figure 3. Pn travel-time residuals versus epicentral distance for station BRVK, before (red squares) and after(green circles) applying model-based Pn SSSCs computed by Bondár’s method.

Figure 4. Pn travel-time residuals versus epicentral distance for station BRVK, before (red squares) and after(green circles) applying model-based plus kriged Pn SSSCs.

of 20

CCB-PRO-02/04 Rev.1

n
Figure 5. Mislocation distances with and without using model-based SSSCs with respect to corresponding GTlocations. The green symbols show the events for which the mislocation error is smaller using SSSCs thawithout. Red symbols show the events for which the mislocation errors are smaller without using SSSCs. Thebisecting line corresponds to equivalent mislocation errors for the two solutions (with and without SSSCs).
100

101

102

100

101

102

Mislocation Distance (with SSSCs) [km]

Mis

loca

tion

Dis

tanc

e (w

ithou

t SS

SC

s) [k

m]

Difference in location from GT

Better BisectorWorse

of 20

CCB-PRO-02/04 Rev.1

Figure 6. Scatter plot of error ellipse areas computed with (x-axis) and without (y-axis) using model-basedSSSCs. Green symbols represent error ellipse areas that are smaller when using the SSSCs than without.

102

103

104

105

102

103

104

105

Area (with SSSCs) (km**2)

Are

a (w

ithou

t SS

SC

s) (

km**

2)

Area of error ellipse

Better Bisector

of 20

CCB-PRO-02/04 Rev.1

Figure 7. Mislocation distances with and without using kriged SSSCs with respect to corresponding GTlocations. Markers and the line are defined as in Figure 5.

10−2

10−1

100

101

102

10−2

10−1

100

101

102

Mislocation Distance (with SSSCs) [km]

Mis

loca

tion

Dis

tanc

e (w

ithou

t SS

SC

s) [k

m]

Difference in location from GT


of 20

CCB-PRO-02/04 Rev.1

Figure 8. Scatter plot of error ellipse areas computed with (x-axis) and without (y-axis) using kriged SSSCs.Markers are defined as in Figure 6.

101

102

103

104

105

106

101

102

103

104

105

106

Area (with SSSCs) (km**2)

Are

a (w

ithou

t SS

SC

s) (

km**

2)

Area of error ellipse


of 20

CCB-PRO-02/04 Rev.1

Figure 9. Kriged surface of Pn travel-time residuals relative to model-based SSSCs for station BRVK.

BRVK Pn Kriged Surface

50˚

60˚

70˚ 80˚

90˚

40˚

50˚

60˚


50˚

60˚

70˚ 80˚

90˚

40˚

50˚

60˚


50˚

60˚

70˚ 80˚

90˚

40˚

50˚

60˚

0 500 1000

km

Above SurfaceBelow Surface

BRVK

-4 -3 -2 -1 0 1 2

seconds

of 20

CCB-PRO-02/04 Rev.1

Figure 10. Kriged surface of Pn travel-time residuals relative to model-based SSSCs for station NRI.

NRI Pn Kriged Surface

60˚

70˚

80˚ 90˚

90˚

100˚

110˚

50˚

60˚


60˚

70˚

80˚ 90˚

90˚

100˚

110˚

50˚

60˚


60˚

70˚

80˚ 90˚

90˚

100˚

110˚

50˚

60˚

0 500 1000

km


NRI

-2 -1 0 1 2

seconds

of 20

CCB-PRO-02/04 Rev.1

Figure 11. Kriged surface of Pn travel-time residuals relative to model-based SSSCs for station YAK.

YAK Pn Kriged Surface

110˚

120˚

130˚

140˚

150˚

40˚

50˚

60˚


110˚

120˚

130˚

140˚

150˚

40˚

50˚

60˚


110˚

120˚

130˚

140˚

150˚

40˚

50˚

60˚

0 500 1000

km


YAK

-2 -1 0 1 2

seconds

of 20

CCB-PRO-02/04 Rev.1

Figure 12. Pn modeling errors as functions of epicentral distance (total path length from event to station) forIASPEI91 (red curve), our regionalized model (green curve), and three types of regions of Northern Eurasia(black curves), defined by Kirichenko and Kraev (2001). Also shown are travel-time misfits to the model-basedSSSCs, binned by distance, for Pn phase arrivals in Kitov’s data set (blue markers).

Pn Modeling Error vs. Distance

0

1

2

3

4

Mo

del

Err

or

(sec

on

ds)

0 2 4 6 8 10 12 14 16 18 20

Epicentral Distance (degrees)

IASPEI91Lamont Model ErrorPlatforms AreasPaleozoic MassifsActive Tectonic AreasTravel-Time Misfit

of 20

CCB-PRO-02/04 Rev.1

Appendix A: Validation Test Report from Columbia University Group 1Consortium

Appendix B: Validation Test Plan

Appendix C: Validation Test Report from RDSS

Appendix D: Recommendations from 1999 Oslo Workshop

of 20

Memo - Columbia University

Documents

Transcript of Memo - Columbia University